Qwen3.5 40B fine-tunes add uncensored variants

This feels less like a polished commercial launch and more like a fast-moving model shop, but the recipe is compelling: scale Qwen3.5 up, distill Claude-style reasoning into it, and hand the community local-friendly quants.

• –The core technical hook is a 40B dense Qwen3.5 variant with 96 layers and 1275 tensors, tuned on Claude 4.6 Opus reasoning data via Unsloth.
• –The variant split matters: regular for general use, Heretic for uncensored output, and RoughHouse for a wilder creative mode. That makes the release more about behavior control than raw benchmark chasing.
• –GGUF availability is the practical win here, because it puts the model in reach of people running llama.cpp-style stacks rather than only GPU-heavy servers.
• –The model card hints at long-context and variable-length reasoning, which is exactly the kind of setup local power users want for writing, roleplay, and code-adjacent tasks.
• –The naming is chaotic, but the collection strategy is coherent: this looks like a repeatable tuning pipeline across the Qwen3.5 family, not a one-off experiment.